On the basis of both adaptive BP algorithm and Newtons method, Quasi Newton algorithm with adaptive decoupled step and momentum (QNADSM) for feed-forward neural networks is derived.
基于输出层函数为线性函数的三层前馈神经网络，结合自适应步长和动量解耦的伪牛顿算法及迭代最小二乘法导出了一种混合算法。
The disadvantages of quasi-Newton algorithm is the great memory, so for large problems, memory difficulties may be encountered.
拟牛顿算法的缺点是所需存储量较大，对于大型问题，可能遇到存储方面的困难。
Compared with the quasi Newton methods and ACA, the solution accuracy of new algorithm is not only improved, but also the convergent reliability is increased.
与拟牛顿法和蚁群算法相比，新算法不仅提高了解的精确性，而且增强了收敛的可靠性。
应用推荐